Maximum Transmission Unit

Maximum Transmission Unit describes a packet-size limit, not a bandwidth limit. On an Ethernet interface, an IP MTU of 1500 bytes normally allows an IP packet containing its IP header and transport-layer data to occupy up to 1500 bytes. Ethernet headers and the frame check sequence are additional: a standard untagged Ethernet frame carrying that packet is 1518 bytes, excluding the preamble and interpacket gap. This distinction matters because switch vendors may express maximum frame size differently from an operating system's IP MTU.

The path MTU is the smallest packet-size limit along the route, accounting for encapsulation where applicable. Configuring a server for a 9001-byte MTU does not make every destination reachable with 9001-byte packets. Virtual interfaces, routers, firewalls, transit networks, tunnels, and destination interfaces can impose smaller limits. Forward and return paths can also differ, so testing only one direction is insufficient.

Jumbo frames carry IP packets larger than the conventional 1500-byte Ethernet MTU. They can reduce packets per second, protocol overhead, and host processing costs for sustained transfers such as replication or storage traffic. However, they require compatible handling throughout the intended path. Larger frames take longer to serialize on a link and are not automatically beneficial for small messages or latency-sensitive workloads. Increasing MTU does not increase the physical link rate or guarantee lower cloud data-transfer charges.

For physical interconnects, verify the supported MTU on the customer router, switch ports, provider connection, and logical interfaces. An 802.1Q VLAN tag adds four bytes to the Ethernet frame; equipment must accommodate that overhead without unintentionally reducing the intended IP MTU. A link aggregation group increases aggregate capacity and can improve resilience, but it does not increase MTU or necessarily increase the throughput of one flow. Compatible optics and link speeds establish the physical connection; they do not prove jumbo-frame support.

Cloud paths frequently mix MTU limits. Supported Amazon EC2 interfaces can use a 9001-byte MTU. AWS Direct Connect supports 9001 bytes on private virtual interfaces and 8500 bytes on transit virtual interfaces, while public virtual interfaces use 1500 bytes. AWS Transit Gateway supports an MTU of 8500 bytes on supported non-VPN attachment paths. Consequently, a 9001-byte EC2 packet cannot cross a transit-gateway path unchanged. AWS Site-to-Site VPN does not support jumbo frames; its maximum MTU is 1446 bytes, and the usable value can be lower depending on the encryption algorithms and encapsulation. Always evaluate the actual route rather than assuming a dedicated connection makes every attached path jumbo-capable.

IPv4 routers can fragment oversized packets when fragmentation is permitted, but packets with the Don't Fragment flag set must instead be dropped and normally trigger an ICMP fragmentation-needed response. IPv6 routers do not fragment forwarded packets; they return ICMPv6 Packet Too Big messages so the sender can adjust. Path MTU Discovery relies on these signals. If a firewall, security group, or network ACL blocks the necessary messages, small requests can succeed while large responses repeatedly retransmit. Some network services also do not provide the feedback required for conventional Path MTU Discovery, so discovery cannot be assumed to work across every cloud path.

When troubleshooting, compare interface MTUs, map the route and its attachment types, and probe progressively larger packets. On Linux, an IPv4 command such as ping -M do -s 1472 DESTINATION tests a 1500-byte IP packet with ordinary 20-byte IPv4 and 8-byte ICMP headers. Failure alone does not prove an MTU mismatch because ICMP echo might be blocked. Correlate tests with packet captures, retransmissions, interface errors, and ICMP responses. Account for segmentation and receive offloads when interpreting captures, because host captures can display packets larger than the frames actually transmitted.

Restore connectivity by correcting an inconsistent link configuration, lowering the source or tunnel MTU, or allowing the required discovery messages where supported. TCP maximum segment size clamping can prevent oversized TCP segments across constrained paths, but it does not fix UDP or replace correct MTU design. Validate both the preferred route and failover routes: a jumbo-capable dedicated connection might fail over to a substantially smaller VPN path.